Predictions of wine ratings using natural language processing

Abstract

In this paper, we built the machine learning models (Decision Tree, Gradient Boosting, and Random Forest) to predict the quality of wines by text-mining the verbal descriptions of sommeliers as well as grape varieties, countries, and sommeliers. We applied the natural language processing to verbal descriptions or commentaries accompanying the wine ratings. We found that verbal descriptions can predict the quality of wines more accurately than the price. Using our models, we believe that wine ratings can represent humans' subjective feelings about wines as well as their objective chemical compositions.